In order to solve over-fitting problem in deep convolution neural network（CNN）

an improved structure is presented. Dropout is used at training time

model averaging method used at test stage

considers retaining probability and probability of each unit within pooling region at the same time.Simulation experiments show that in MNIST handwritten database and part of CMU-PIE database

under the same structure and the same number of iterations

the method is superior to other methods.It can get a better recognition rate and can overcome over-fitting problem better.